feat: Add streaming to HuggingFaceLocalGenerator

[vblagoje]

Why:

The change introduces a new streaming_callback parameter to the HuggingFaceLocalGenerator class, allowing users to handle streaming responses. This enhancement provides greater flexibility and customization in managing real-time responses from Hugging Face's local models.

• fixes HuggingFaceLocalGenerator - streaming/stream handler support #6048

What:

1. Modifies haystack/components/generators/hugging_face_local.py:

   • Adds streaming_callback as an optional parameter to the HuggingFaceLocalGenerator class.
   • Implements streaming_callback functionality within the run method.
   • Updates to_dict and from_dict methods to serialize and deserialize the streaming_callback parameter.

2. Adds releasenotes/notes/hugging-face-local-generator-streaming-callback-38a77d37199f9672.yaml to document the new feature.

3. Modifies test/components/generators/test_hugging_face_local_generator.py to include tests for the streaming_callback parameter.

How can it be used:

1. Include streaming_callback when initializing a HuggingFaceLocalGenerator instance to handle streaming responses:

   generator = HuggingFaceLocalGenerator(model="model_name", streaming_callback=streaming_callback_handler)

2. Implement the streaming_callback function to handle streaming responses as desired, for example:

   def streaming_callback_handler(x):
       print(x)

How did you test it:

1. Tested the streaming_callback by creating tests for:
   • Default behavior (no streaming callback)
   • Serialization and deserialization (to_dict and from_dict methods)
   • Custom streaming callback function implementation
   • Interaction with the run method

Notes for the reviewer:

• Ensure that the streaming_callback function is compatible with the async context of the HuggingFaceLocalGenerator's run method.
• Test the streaming feature with local Hugging Face models to validate the behavior.
• Test the serialization and deserialization functionality of the streaming_callback parameter.
• Consider adding integration tests for various Hugging Face models to ensure compatibility with different types of streaming data.

[coveralls]

Pull Request Test Coverage Report for Build 8375520678

Details

• 0 of 0 changed or added relevant lines in 0 files are covered.
• 6 unchanged lines in 1 file lost coverage.
• Overall coverage increased (+0.02%) to 89.28%

---

Files with Coverage Reduction	New Missed Lines	%
components/generators/hugging_face_local.py	6	91.89%

Totals
Change from base Build 8374477800:	0.02%
Covered Lines:	5422
Relevant Lines:	6073

---

💛 - Coveralls

[davidsbatista]

I think it makes more sense to define this inside the test_run_with_streaming(self) itself, since it's only used there

[vblagoje]

Good suggestion @davidsbatista - I would but the serialization mechanism doesn't see the namespace of the function IIRC.

[vblagoje]

Nevermind, it can be done, thanks for the suggestion

[davidsbatista]

just a small suggestions, other than that LGTM

[vblagoje]

@davidsbatista rebased to latest main, resolved a few conflicts, should be ready to go now, just waiting for green signal from CI